Table of Contents

1  Ground predictions
1.1  PVLib Clearsky
2  Train/test on NSRDB data to find optimal parameters
3  Train on all NSRDB data, test various freq of ground data
3.1  30 min freq ground data
3.2  15 min freq ground data
3.3  10 min freq ground data
3.4  5 min freq ground data
3.5  1 min freq ground data
4  Save model
5  Conclusion

In [1]:
import pandas as pd
import numpy as np
import os
import datetime
import matplotlib
import matplotlib.pyplot as plt
from mpl_toolkits.axes_grid1 import make_axes_locatable
from sklearn import tree
from sklearn import ensemble
from sklearn import linear_model

import pytz
import itertools
import visualize
import utils
import pydotplus
import xgboost as xgb

from sklearn import metrics
from sklearn import model_selection

import pvlib
import cs_detection

import visualize_plotly as visualize
from plotly.offline import download_plotlyjs, init_notebook_mode, plot, iplot
import plotly.graph_objs as go
init_notebook_mode(connected=True)

from IPython.display import Image

%load_ext autoreload
%autoreload 2

np.set_printoptions(precision=4)
%matplotlib notebook

Ground predictions

PVLib Clearsky

Only making ground predictions using PVLib clearsky model and statistical model. NSRDB model won't be available to ground measurements.

In [139]:
nsrdb = cs_detection.ClearskyDetection.read_pickle('abq_nsrdb_1.pkl.gz')
nsrdb.df.index = nsrdb.df.index.tz_convert('MST')
In [143]:
nsrdb.df[nsrdb.df['GHI'] > 0]['sky_status'].value_counts()
Out[143]:
True     82430
False    73208
Name: sky_status, dtype: int64
nsrdb.trim_dates('01-01-2014', '01-01-2016')
In [3]:
nsrdb.time_from_solar_noon('Clearsky GHI pvlib', 'tfn')
nsrdb.time_from_solar_noon_ratio2('Clearsky GHI pvlib')
In [4]:
feature_cols = [
'ghi_status',
'tfn',
'abs_ideal_ratio_diff',
'abs_ideal_ratio_diff mean',
'abs_ideal_ratio_diff std',
'abs_ideal_ratio_diff max',
'abs_ideal_ratio_diff min',
'GHI Clearsky GHI pvlib gradient ratio', 
'GHI Clearsky GHI pvlib gradient ratio mean', 
'GHI Clearsky GHI pvlib gradient ratio std', 
'GHI Clearsky GHI pvlib gradient ratio min', 
'GHI Clearsky GHI pvlib gradient ratio max', 
'GHI Clearsky GHI pvlib gradient second ratio', 
'GHI Clearsky GHI pvlib gradient second ratio mean', 
'GHI Clearsky GHI pvlib gradient second ratio std', 
'GHI Clearsky GHI pvlib gradient second ratio min', 
'GHI Clearsky GHI pvlib gradient second ratio max', 
'GHI Clearsky GHI pvlib line length ratio',
'GHI Clearsky GHI pvlib line length ratio gradient',
'GHI Clearsky GHI pvlib line length ratio gradient second'
]

target_cols = ['sky_status']

Train/test on NSRDB data to find optimal parameters

In [73]:
train = cs_detection.ClearskyDetection(nsrdb.df)
train.trim_dates(None, '01-01-2015')
test = cs_detection.ClearskyDetection(nsrdb.df)
test.trim_dates('01-01-2015', None)
In [74]:
train.scale_model('GHI', 'Clearsky GHI pvlib', 'sky_status')
train.time_from_solar_noon_ratio2('Clearsky GHI pvlib')
train.scale_by_irrad('Clearsky GHI pvlib')
In [7]:
params={}
params['C'] = [.001, .01, .1, 1, 10]
params['penalty'] = ['l1', 'l2']
# params['n_estimators'] = [32, 64, 128, 256]
best_score = -1
for c, p in itertools.product(params['C'], params['penalty']):
    train2 = cs_detection.ClearskyDetection(train.df)
    train2.trim_dates('01-01-1999', '01-01-2014')
    utils.calc_all_window_metrics(train2.df, 3, meas_col='GHI', model_col='Clearsky GHI pvlib', overwrite=True)
    test2 = cs_detection.ClearskyDetection(train.df)
    test2.trim_dates('01-01-2014', '01-01-2015')
    clf = linear_model.LogisticRegression(C=c, penalty=p)
    clf.fit(train2.df[feature_cols].values, train2.df[target_cols].values.flatten())
    pred = test2.iter_predict_daily(feature_cols, 'GHI', 'Clearsky GHI pvlib', clf, 3, multiproc=True, by_day=True).astype(bool)
    f1_score = metrics.f1_score(test2.df['sky_status'], pred)
    recall_score = metrics.recall_score(test2.df['sky_status'], pred)
    precision_score = metrics.precision_score(test2.df['sky_status'], pred)
#     if score > best_score:
#         best_params = {}
#         best_params['max_depth'] = depth
#         best_params['n_estimators'] = nest
#         best_score = score
    print(f1_score, recall_score, precision_score, c, p)
# print(score, best_params)

0.867968073853 0.938839192844 0.807045779685 0.001 l1

/Users/benellis/miniconda3/lib/python3.5/site-packages/sklearn/metrics/classification.py:1113: UndefinedMetricWarning:

F-score is ill-defined and being set to 0.0 due to no predicted samples.

/Users/benellis/miniconda3/lib/python3.5/site-packages/sklearn/metrics/classification.py:1113: UndefinedMetricWarning:

Precision is ill-defined and being set to 0.0 due to no predicted samples.


0.0 0.0 0.0 0.001 l2
0.892016968762 0.962346577907 0.831266846361 0.01 l1
0.0 0.0 0.0 0.01 l2
0.905546935609 0.971291866029 0.848138056312 0.1 l1

/Users/benellis/miniconda3/lib/python3.5/site-packages/sklearn/metrics/classification.py:1113: UndefinedMetricWarning:

F-score is ill-defined and being set to 0.0 due to no predicted samples.

/Users/benellis/miniconda3/lib/python3.5/site-packages/sklearn/metrics/classification.py:1113: UndefinedMetricWarning:

Precision is ill-defined and being set to 0.0 due to no predicted samples.


0.0 0.0 0.0 0.1 l2
0.907855131566 0.972540045767 0.851238164603 1 l1

/Users/benellis/miniconda3/lib/python3.5/site-packages/sklearn/metrics/classification.py:1113: UndefinedMetricWarning:

F-score is ill-defined and being set to 0.0 due to no predicted samples.

/Users/benellis/miniconda3/lib/python3.5/site-packages/sklearn/metrics/classification.py:1113: UndefinedMetricWarning:

Precision is ill-defined and being set to 0.0 due to no predicted samples.


0.0 0.0 0.0 1 l2
0.909815354713 0.973788225504 0.853729710013 10 l1
0.0 0.0 0.0 10 l2

/Users/benellis/miniconda3/lib/python3.5/site-packages/sklearn/metrics/classification.py:1113: UndefinedMetricWarning:

F-score is ill-defined and being set to 0.0 due to no predicted samples.

/Users/benellis/miniconda3/lib/python3.5/site-packages/sklearn/metrics/classification.py:1113: UndefinedMetricWarning:

Precision is ill-defined and being set to 0.0 due to no predicted samples.
In [ ]:
In [ ]:
In [ ]:
In [ ]:
In [75]:
utils.calc_all_window_metrics(train.df, 3, meas_col='GHI', model_col='Clearsky GHI pvlib', overwrite=True)
In [76]:
# f1
best_params = {'penalty': 'l1', 'C':10, 'class_weight': 'balanced'}
# recall
# same as f1
# best_params = {'penalty': 'l1', 'C':}
# precision
# same as f1
In [77]:
clf = linear_model.LogisticRegression(**best_params)
clf.fit(train.df[feature_cols].values, train.df[target_cols].values.flatten())
Out[77]:
LogisticRegression(C=10, class_weight='balanced', dual=False,
          fit_intercept=True, intercept_scaling=1, max_iter=100,
          multi_class='ovr', n_jobs=1, penalty='l1', random_state=None,
          solver='liblinear', tol=0.0001, verbose=0, warm_start=False)
In [78]:
test = cs_detection.ClearskyDetection(nsrdb.df)
test.trim_dates('01-01-2015', None)
test.scale_by_irrad('Clearsky GHI pvlib')
In [79]:
%%time
pred = test.iter_predict_daily(feature_cols, 'GHI', 'Clearsky GHI pvlib', clf, 3, multiproc=True, by_day=True).astype(bool)

CPU times: user 1.27 s, sys: 194 ms, total: 1.47 s
Wall time: 32.5 s
In [80]:
vis = visualize.Visualizer()
vis.add_line_ser(test.df['GHI'], 'GHI')
vis.add_line_ser(test.df['Clearsky GHI pvlib'], 'GHI_cs')
vis.add_circle_ser(test.df[(test.df['sky_status'] == 0) & (pred)]['GHI'], 'ML clear only')
vis.add_circle_ser(test.df[(test.df['sky_status'] == 1) & (~pred)]['GHI'], 'NSRDB clear only')
vis.add_circle_ser(test.df[(test.df['sky_status'] == 1) & (pred)]['GHI'], 'ML+NSRDB clear only')
vis.show()
In [81]:
cm = metrics.confusion_matrix(test.df['sky_status'].values, pred)
vis = visualize.Visualizer()
vis.plot_confusion_matrix(cm, labels=['cloudy', 'clear'])
In [82]:
print(metrics.f1_score(test.df['sky_status'].values, pred))

0.898556535686

Train on all NSRDB data, test various freq of ground data

In [83]:
train = cs_detection.ClearskyDetection(nsrdb.df)
train.scale_model('GHI', 'Clearsky GHI pvlib', 'sky_status')
train.scale_by_irrad('Clearsky GHI pvlib')
utils.calc_all_window_metrics(train.df, 3, meas_col='GHI', model_col='Clearsky GHI pvlib', overwrite=True)
clf.fit(train.df[feature_cols].values, train.df[target_cols].values.flatten())
Out[83]:
LogisticRegression(C=10, class_weight='balanced', dual=False,
          fit_intercept=True, intercept_scaling=1, max_iter=100,
          multi_class='ovr', n_jobs=1, penalty='l1', random_state=None,
          solver='liblinear', tol=0.0001, verbose=0, warm_start=False)
In [84]:
clf.coef_
Out[84]:
array([[  1.2059e+01,  -2.5510e-04,  -2.1303e+02,  -2.4385e+01,
          1.4682e+01,  -1.0641e+01,   2.2135e+02,   1.0804e-04,
         -2.0488e-02,   1.0017e-02,  -4.0609e-02,   8.0827e-04,
         -8.8661e-04,  -1.8208e-02,   7.3477e-03,   2.1432e-03,
          1.7840e-03,   5.9074e-01,   3.3273e+00,  -1.5647e+02]])
In [85]:
bar = go.Bar(x=feature_cols, y=clf.coef_)
iplot([bar])

30 min freq ground data

In [86]:
ground = cs_detection.ClearskyDetection.read_pickle('abq_ground_1.pkl.gz')
ground.df.index = ground.df.index.tz_convert('MST')
test = cs_detection.ClearskyDetection(ground.df)
In [87]:
test.trim_dates('10-01-2015', '10-08-2015')
In [88]:
test.time_from_solar_noon('Clearsky GHI pvlib', 'tfn')
test.time_from_solar_noon_ratio2('Clearsky GHI pvlib')
test.scale_by_irrad('Clearsky GHI pvlib')
In [89]:
test.df = test.df[test.df.index.minute % 30 == 0]
In [90]:
# test.df.loc[np.round(test.df['GHI'], 6) == 14.48218, 'GHI'] = 40
In [91]:
# test.df = test.df.resample('30T').apply(lambda x: x[len(x) // 2])
In [92]:
pred = test.iter_predict_daily(feature_cols, 'GHI', 'Clearsky GHI pvlib', clf, 3, multiproc=True, by_day=True).astype(bool)
In [93]:
train2 = cs_detection.ClearskyDetection(nsrdb.df)
train2.intersection(test.df.index)
In [94]:
nsrdb_clear = train2.df['sky_status'].values
ml_clear = pred
vis = visualize.Visualizer()
vis.add_line_ser(test.df['GHI'], 'GHI')
vis.add_line_ser(test.df['Clearsky GHI pvlib'], 'GHI_cs')
vis.add_circle_ser(test.df[ml_clear & ~nsrdb_clear]['GHI'], 'ML clear only')
vis.add_circle_ser(test.df[~ml_clear & nsrdb_clear]['GHI'], 'NSRDB clear only')
vis.add_circle_ser(test.df[ml_clear & nsrdb_clear]['GHI'], 'Both clear')
# vis.add_line_ser(test.df['abs_ideal_ratio_diff'] * 100)
# vis.add_line_ser(test.df['GHI Clearsky GHI pvlib line length ratio'] * 100)
# vis.add_line_ser(test.df['GHI Clearsky GHI pvlib gradient ratio'] * 100)
# vis.add_line_ser(test.df['GHI Clearsky GHI pvlib gradient second ratio'] * 100)
# vis.add_line_ser(test.df['GHI Clearsky GHI pvlib gradient ratio std'] * 100)
# vis.add_line_ser(test.df['irrad_scaler'] * 100)
vis.show()
In [95]:
probas = clf.predict_proba(test.df[feature_cols].values)
In [96]:
test.df['probas'] = 0
In [97]:
test.df['probas'] = probas
In [98]:
trace0 = go.Scatter(x=test.df.index, y=test.df['GHI'], name='GHI')
trace1 = go.Scatter(x=test.df.index, y=test.df['Clearsky GHI pvlib'], name='GHIcs')
trace2 = go.Scatter(x=test.df.index, y=test.df['GHI'], name='prob', mode='markers', marker={'size': 12, 'color': test.df['probas'], 'colorscale': 'Viridis', 'showscale': True}, text='prob: ' + test.df['probas'].astype(str))
iplot([trace0, trace1, trace2])
In [99]:
visualize.plot_ts_slider_highligther(test.df, prob='probas')

15 min freq ground data

In [100]:
ground = cs_detection.ClearskyDetection.read_pickle('abq_ground_1.pkl.gz')
ground.df.index = ground.df.index.tz_convert('MST')
test = cs_detection.ClearskyDetection(ground.df)
In [101]:
test.trim_dates('10-01-2015', '10-08-2015')
In [102]:
test.time_from_solar_noon('Clearsky GHI pvlib', 'tfn')
test.scale_by_irrad('Clearsky GHI pvlib')
In [103]:
test.df = test.df[test.df.index.minute % 15 == 0]
# test.df = test.df.resample('15T').apply(lambda x: x[len(x) // 2])
In [104]:
pred = test.iter_predict_daily(feature_cols, 'GHI', 'Clearsky GHI pvlib', clf, 5, multiproc=True, by_day=True).astype(bool)
In [105]:
train2 = cs_detection.ClearskyDetection(train.df)
train2.trim_dates('10-01-2015', '10-08-2015')
train2.df = train2.df.reindex(pd.date_range(start=train2.df.index[0], end=train2.df.index[-1], freq='15min'))
train2.df['sky_status'] = train2.df['sky_status'].fillna(False)
In [106]:
nsrdb_clear = train2.df['sky_status']
ml_clear = test.df['sky_status iter']
vis = visualize.Visualizer()
vis.add_line_ser(test.df['GHI'], 'GHI')
vis.add_line_ser(test.df['Clearsky GHI pvlib'], 'GHI_cs')
vis.add_circle_ser(test.df[ml_clear & ~nsrdb_clear]['GHI'], 'ML clear only')
vis.add_circle_ser(test.df[~ml_clear & nsrdb_clear]['GHI'], 'NSRDB clear only')
vis.add_circle_ser(test.df[ml_clear & nsrdb_clear]['GHI'], 'Both clear')
vis.show()
In [107]:
probas = clf.predict_proba(test.df[feature_cols].values)
test.df['probas'] = 0
test.df['probas'] = probas

trace0 = go.Scatter(x=test.df.index, y=test.df['GHI'], name='GHI')
trace1 = go.Scatter(x=test.df.index, y=test.df['Clearsky GHI pvlib'], name='GHIcs')
trace2 = go.Scatter(x=test.df.index, y=test.df['GHI'], name='prob', mode='markers', marker={'size': 12, 'color': test.df['probas'], 'colorscale': 'Viridis', 'showscale': True}, text=test.df['probas'])
iplot([trace0, trace1, trace2])
In [108]:
visualize.plot_ts_slider_highligther(test.df, prob='probas')
In [ ]:

10 min freq ground data

In [109]:
ground = cs_detection.ClearskyDetection.read_pickle('abq_ground_1.pkl.gz')
ground.df.index = ground.df.index.tz_convert('MST')
test = cs_detection.ClearskyDetection(ground.df)
In [110]:
test.trim_dates('10-01-2015', '10-08-2015')
In [111]:
test.time_from_solar_noon('Clearsky GHI pvlib', 'tfn')
test.scale_by_irrad('Clearsky GHI pvlib')
In [112]:
test.df = test.df[test.df.index.minute % 10 == 0]
In [113]:
pred = test.iter_predict_daily(feature_cols, 'GHI', 'Clearsky GHI pvlib', clf, 7, multiproc=True, by_day=True).astype(bool)
In [114]:
train2 = cs_detection.ClearskyDetection(train.df)
train2.trim_dates('10-01-2015', '10-08-2015')
train2.df = train2.df.reindex(pd.date_range(start=train2.df.index[0], end=train2.df.index[-1], freq='10min'))
train2.df['sky_status'] = train2.df['sky_status'].fillna(False)
In [115]:
nsrdb_clear = train2.df['sky_status']
ml_clear = test.df['sky_status iter']
vis = visualize.Visualizer()
vis.add_line_ser(test.df['GHI'], 'GHI')
vis.add_line_ser(test.df['Clearsky GHI pvlib'], 'GHI_cs')
vis.add_circle_ser(test.df[ml_clear & ~nsrdb_clear]['GHI'], 'ML clear only')
vis.add_circle_ser(test.df[~ml_clear & nsrdb_clear]['GHI'], 'NSRDB clear only')
vis.add_circle_ser(test.df[ml_clear & nsrdb_clear]['GHI'], 'Both clear')
vis.show()
In [116]:
probas = clf.predict_proba(test.df[feature_cols].values)
test.df['probas'] = 0
test.df['probas'] = probas

trace0 = go.Scatter(x=test.df.index, y=test.df['GHI'], name='GHI')
trace1 = go.Scatter(x=test.df.index, y=test.df['Clearsky GHI pvlib'], name='GHIcs')
trace2 = go.Scatter(x=test.df.index, y=test.df['GHI'], name='prob', mode='markers', marker={'size': 12, 'color': test.df['probas'], 'colorscale': 'Viridis', 'showscale': True}, text=test.df['probas'])
iplot([trace0, trace1, trace2])
In [117]:
visualize.plot_ts_slider_highligther(test.df, prob='probas')

5 min freq ground data

In [118]:
ground = cs_detection.ClearskyDetection.read_pickle('abq_ground_1.pkl.gz')
ground.df.index = ground.df.index.tz_convert('MST')
test = cs_detection.ClearskyDetection(ground.df)
In [119]:
test.trim_dates('10-01-2015', '10-08-2015')
In [120]:
test.time_from_solar_noon('Clearsky GHI pvlib', 'tfn')
test.scale_by_irrad('Clearsky GHI pvlib')
In [121]:
test.df = test.df[test.df.index.minute % 5 == 0]
In [122]:
pred = test.iter_predict_daily(feature_cols, 'GHI', 'Clearsky GHI pvlib', clf, 13, multiproc=True, by_day=True).astype(bool)
In [123]:
train2 = cs_detection.ClearskyDetection(train.df)
train2.trim_dates('10-01-2015', '10-08-2015')
train2.df = train2.df.reindex(pd.date_range(start=train2.df.index[0], end=train2.df.index[-1], freq='5min'))
train2.df['sky_status'] = train2.df['sky_status'].fillna(False)
In [124]:
nsrdb_clear = train2.df['sky_status']
ml_clear = test.df['sky_status iter']
vis = visualize.Visualizer()
vis.add_line_ser(test.df['GHI'], 'GHI')
vis.add_line_ser(test.df['Clearsky GHI pvlib'], 'GHI_cs')
vis.add_circle_ser(test.df[ml_clear & ~nsrdb_clear]['GHI'], 'ML clear only')
vis.add_circle_ser(test.df[~ml_clear & nsrdb_clear]['GHI'], 'NSRDB clear only')
vis.add_circle_ser(test.df[ml_clear & nsrdb_clear]['GHI'], 'Both clear')
vis.show()
In [125]:
probas = clf.predict_proba(test.df[feature_cols].values)
test.df['probas'] = 0
test.df['probas'] = probas

trace0 = go.Scatter(x=test.df.index, y=test.df['GHI'], name='GHI')
trace1 = go.Scatter(x=test.df.index, y=test.df['Clearsky GHI pvlib'], name='GHIcs')
trace2 = go.Scatter(x=test.df.index, y=test.df['GHI'], name='prob', mode='markers', marker={'size': 10, 'color': test.df['probas'], 'colorscale': 'Viridis', 'showscale': True}, text=test.df['probas'])
iplot([trace0, trace1, trace2])
In [126]:
visualize.plot_ts_slider_highligther(test.df, prob='probas')

1 min freq ground data

In [127]:
ground = cs_detection.ClearskyDetection.read_pickle('abq_ground_1.pkl.gz')
ground.df.index = ground.df.index.tz_convert('MST')
test = cs_detection.ClearskyDetection(ground.df)
In [128]:
test.trim_dates('10-01-2015', '10-08-2015')
In [129]:
test.time_from_solar_noon('Clearsky GHI pvlib', 'tfn')
test.scale_by_irrad('Clearsky GHI pvlib')
In [130]:
test.df = test.df[test.df.index.minute % 1 == 0]
In [131]:
pred = test.iter_predict_daily(feature_cols, 'GHI', 'Clearsky GHI pvlib', clf, 61, multiproc=True, by_day=True).astype(bool)
In [132]:
train2 = cs_detection.ClearskyDetection(train.df)
train2.trim_dates('10-01-2015', '10-08-2015')
train2.df = train2.df.reindex(pd.date_range(start=train2.df.index[0], end=train2.df.index[-1], freq='1min'))
train2.df['sky_status'] = train2.df['sky_status'].fillna(False)
In [133]:
nsrdb_clear = train2.df['sky_status']
ml_clear = test.df['sky_status iter']
vis = visualize.Visualizer()
vis.add_line_ser(test.df['GHI'], 'GHI')
vis.add_line_ser(test.df['Clearsky GHI pvlib'], 'GHI_cs')
vis.add_circle_ser(test.df[ml_clear & ~nsrdb_clear]['GHI'], 'ML clear only')
vis.add_circle_ser(test.df[~ml_clear & nsrdb_clear]['GHI'], 'NSRDB clear only')
vis.add_circle_ser(test.df[ml_clear & nsrdb_clear]['GHI'], 'Both clear')
vis.show()
In [134]:
probas = clf.predict_proba(test.df[feature_cols].values)
test.df['probas'] = 0
test.df['probas'] = probas

trace0 = go.Scatter(x=test.df.index, y=test.df['GHI'], name='GHI')
trace1 = go.Scatter(x=test.df.index, y=test.df['Clearsky GHI pvlib'], name='GHIcs')
trace2 = go.Scatter(x=test.df.index, y=test.df['GHI'], name='prob', mode='markers', marker={'size': 10, 'color': test.df['probas'], 'colorscale': 'Viridis', 'showscale': True}, text=test.df['probas'])
iplot([trace0, trace1, trace2])

/Users/benellis/miniconda3/lib/python3.5/site-packages/sklearn/linear_model/base.py:352: RuntimeWarning:

overflow encountered in exp
In [135]:
visualize.plot_ts_slider_highligther(test.df, prob='probas')

Save model

In [136]:
import pickle
In [137]:
with open('abq_trained.pkl', 'wb') as f:
    pickle.dump(clf, f)
In [138]:
!ls abq*

abq_ground.pkl.gz   abq_nsrdb.pkl.gz    abq_nsrdb_1.pkl.gz
abq_ground_1.pkl.gz abq_nsrdb_1.pkl     abq_trained.pkl

Conclusion

In general, the clear sky identification looks good. At lower frequencies (30 min, 15 min) we see good agreement with NSRDB labeled points. I suspect this could be further improved my doing a larger hyperparameter search, or even doing some feature extraction/reduction/additions.

In [ ]: